Speech Recognition For The Contact Center
By Ken Waln, Edify Corporation
In today's competitive marketplace, contact centers everywhere are
installing speech recognition systems to improve their responsiveness to
customers. Should you deploy a speech solution in your contact center? The
following questions may help you decide.
' Do callers wait in long queues to perform simple transactions such as
address changes?
' Do your agents spend valuable time identifying callers, an-swering
simple questions and routing calls when they should be providing greater
value to customers?
' Do your customers call while driving, so they have to navigate through
touch-tone menus with one hand, without looking at their matchbox-sized
phones?
' Are you looking for new ways to differentiate your business from your
competitors, retain your existing customers and acquire new ones?
If you answered 'Yes' to any of these questions, it's time to deploy
speech recognition technologies in your contact center.
Although speech technologies have been available for several years and
have made significant inroads in several markets, most contact centers today
use simple touch-tone-based voice response systems for the majority of their
calls. With the advent of standards like VoiceXML and SALT, and support from
major vendors like Microsoft and IBM, speech recognition is quickly moving
into the mainstream. Speech systems are poised to bring significant return
on investment to contact centers through increased automation of customer
calls, partial automation of agent-assisted calls and more accurate call
routing within the contact center.
Along with cost savings, speech systems can bring return on investment in
less tangible ways. By making your company more accessible, these systems
can increase your customers' satisfaction with your service offerings. A
well-designed speech system improves brand awareness just as a good Web site
does. If you do not take advantage of the benefits of speech, your
competitors probably will, and will entice your customers to switch.
Improving Voice Automation
Until now, most speech applications have been deployed in areas where
touch-tone systems simply cannot function. The first deployments were in
industries like financial investment and travel, where information such as
stock symbols or dates, times and destinations were far too complex to enter
onto a keypad. Following the tremendous success of these initial speech
applications, speech automation is rapidly being adopted in a broad range of
industries.
Studies show that callers are much more satisfied by speech-enabled
systems and less likely to opt-out to an operator. In a recent Harris
Interactive Research survey, nine out of ten consumers said they would
derive greater value from speech systems, and 70 percent of users said their
experience would be improved by using speech systems versus touch-tone
systems1. Several factors are driving this increased acceptance, including
improved voice interfaces and the increased difficulty of using keypads on
many contemporary phones, especially cell phones.
To speech-enable your current applications, it is important to recognize
the differences between touch-tone systems and speech. For input, touch-tone
systems rely on the 12 keys of the phone keypad, whereas speech systems can
accommodate menus with hundreds or even thousands of options. In the early
days, speech systems replicated touch-tone options with prompts such as
'Press or say 1.' These early systems were effective, but they did not
significantly increase automation. Instead of reducing the number and levels
of menus, they actually resulted in longer and more tedious prompts.
To benefit fully from the capabilities of speech, companies will need to
replace the user interface of their touch-tone systems with a new voice user
interface (VUI). A good VUI design looks less like a simple flowchart and
more like a complex process diagram, allowing advanced users to skip several
steps forward at a time, while guiding less sophisticated users through the
process.
The real benefits come when you flatten the menu structure, allowing
callers to quickly navigate to where they need to go. To do that, a company
needs to change the way it asks for information. Since our short-term memory
can handle only a few items at a time, long lists of options should be
avoided. Instead, companies should ask callers a general question such as,
'What product are you calling about?' or list a moderate number of choices
and enable the caller to interrupt when they hear the option they want.
Help prompts and simple decision trees can also be used to guide callers
to their desired results. More sophisticated applications use open-ended
prompts such as, 'Briefly describe the reason you are calling today.' These
advanced applications employ open grammars to statistically infer the
purpose of the call. This technique is most practical in call routing
applications, which will be discussed in greater detail below.
Callers tend to interact with speech systems much as they do with live
agents, so it is important that a VUI have the appropriate personality and
tone. Selecting the persona for an application should be done with the same
care that a company would take in choosing a spokesperson for its products.
The persona of a speech system should enhance a company's brand identity and
put callers at ease. In exchange, callers will use the system to its fullest
capabilities, thus freeing live agents for more strategic tasks.
Designing a good VUI requires both upfront planning and continual
refinement. Designers often start by evaluating inbound calls. They note the
way customers ask questions and how they phrase their answers. Then they
test various alternatives with a prototype or simulated system. Prior to
system implementation, a series of pilot tests and tuning cycles is
conducted.
The result is an optimal system that meets all of a company's callers'
needs while avoiding the frustrations that stem from confusing questions and
misconstrued answers. The entire design process may seem daunting, but a
system that is well designed and properly tuned will more quickly reach its
potential and maximize the return on a company's investment.
New Automation Opportunities
Although speech-enabling existing automation systems can yield
significant cost savings, new applications can extend automation to more
complex transactions and create even greater operational efficiencies. With
current speech technologies, it is now possible to automate activities that
touch-tone applications simply cannot address. Product availability lookups,
literature requests, address change and capture and a variety of other
common transactions are ill-suited to keypad input.
With a speech system, however, these tasks become quite practical. In
looking for new automation opportunities, a company should examine its call
volumes and look for high-volume, highly repetitive interactions. In many
environments, these can be largely automated with a fairly standard speech
application.
Another significant emerging application is caller identification. Using
speech print analysis or speaker verification ' a function your live agents
cannot provide ' the system can compare the characteristics of a caller's
voice with previously stored information and determine if the caller is
indeed the person he or she claims to be. Similar to a fingerprint or
retinal scan, these systems use biometric data to prevent fraud and simplify
access to the system. In contact centers in which it is practical to enroll
customers with a voiceprint, this application can provide an added level of
security and reduce the need for hard-to-remember personal identification
numbers.
Improving Call Routing
Most contact centers use touch-tone systems for routing calls to the
correct queue, verifying caller identity and collecting other information up
front so the agent does not have to spend time on these tasks. Speech can
improve the accuracy of this process and enable it to go further, reducing
call length and improving service. The use of touch-tone input to route
calls is very successful in simple call centers with only a few obvious
choices. If the caller wants a replacement part, for example, should he or
she contact sales or service? If the caller wants an airline seat upgrade,
is that domestic reservations or the frequent-flyer service desk? Does a
digital photo printer fall under the category of personal electronics or
computer equipment?
Since touch-tone menus in these situations are often hard to navigate, a
large percentage of calls must be transferred from one agent to another.
When the number of queues is large, a deep menu with lots of choices may be
needed. These choices tend to frustrate callers. Not only are these menus
slow and cumbersome, but the customer often reaches the end without finding
what he or she was looking for. Next time, this customer may simply ignore
the menus and 'zero out' to an operator. All of these factors combine to
make call routing in a complex call center a significant cost factor.
Fortunately, speech systems provide a much better approach. By asking one
or more open-ended questions, then statistically processing the answers, it
is possible to achieve a very high rate of correct routing with a very flat
menu. For example, the system might say, 'Briefly describe the purpose of
your call,' and based on the caller's response, route the call among dozens
of queues with an accuracy of 80 percent or more. The technology used to
achieve this is called statistical language modeling, and the grammars
developed are often called open grammars. These grammars rely on statistical
analysis of the users' input rather than matching every input to a
pre-determined set of responses, and can therefore understand a much wider
variety of inputs and provide a probable interpretation.
Getting Started
Speech technologies can improve a company's operations in so many ways that
it can be difficult to decide what to do first. There is often a trade-off
between doing a strategic contact center redesign and a tactical add-on to
existing systems. When planning for a speech rollout, it is important for
companies to remember that while there is a cost for implementing the
solution, there is also a cost for not doing so.
Once a company determines, for example, that an address change
application could save it $40,000 per month, the company's management should
consider that amount as a cost for every month they delay implementing the
solution. By rolling this application out quickly as a first phase of a
larger project, a company could potentially save enough money to roll out a
more complex phase later. We must, of course, acknowledge and address the
realities of corporate guidelines and budgets, but we should keep in mind
that implementing a system with a high return on investment in a short
period is a good way to prove the technology works.
This kind of analysis usually shows that a phased project with a high ROI
in the early phases makes more sense than a large, 'big-bang' project.
However, it is also important to have the big picture in mind, so the system
and vendors chosen for the first phase can also deliver the subsequent
phases.
Speech systems cannot solve all of a business' problems, but they can
bring great benefits and a very quick return on a company's investment,
improving business and increasing competitive position.
The author is the chief technology officer of Edify Corporation (www.edify.com).
2 Harris Interactive Research study, March 2003.
If you are interested in purchasing reprints of this article (in either
print or HTML format), please visit Reprint Management Services online at
www.reprintbuyer.com or contact a
representative via e-mail at
[email protected] or by phone at 800-290-5460.
For information and subscriptions, visit www.TMCnet.com or
call 203-852-6800.
[
Return
To The May 2004 Table Of Contents ] |